JPH0627715B2 - Quality evaluation device for slab section - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial field of application] The present invention is for evaluating the quality of a steel material, for example, a slab produced by continuous casting, for the presence or absence of defects, the type of defects, the degree of defects, etc. Relating to a device for identifying.

[Prior Art] As a defect that occurs in a cast piece, center segregation, V segregation, internal cracking and the like are known. In general, segregation causes deviation in the structure of the steel material due to the difference in the solidification timing of each part of the steel material when the molten metal solidifies. Specifically, the center segregation is a phenomenon in which impurities gather in the center of a slab that slowly solidifies, and the V segregation appears as an inclination around the center of the slab in which the impurities are V-shaped as a whole. This is a visible phenomenon. Internal cracks are cracks that occur inside the slab, and in continuous casting, cracks tend to occur in the thickness direction of the slab.

In order to evaluate the quality of the slab, it is necessary to identify the various defects as described above. Therefore, conventionally, when performing this kind of quality evaluation, a recorded image showing the distribution of impurities, that is, a sulfur print is created from a cross section obtained by cutting a part of a slab by a known sulfur print method, and this is prepared by an expert. Is generally examined and evaluated by visual inspection.

However, in the conventional inspection by human sense, an accurate evaluation result cannot be obtained because of ambiguity, an expert is required, and a long work time is required.

For this reason, there have been various attempts to automatically perform this kind of quality evaluation by a computer using an image processing method, but in reality, it is known that satisfactory quality evaluation results can be obtained. Not not.

As a partial technique, a technique disclosed in Japanese Patent Laid-Open No. 63-19543 is known as a threshold value setting method for obtaining binarized image information from multi-value gradation image information.

[Problems to be Solved by the Invention] A first object of the present invention is to improve the accuracy of inspection by automatically performing quality evaluation of steel materials without depending on human senses.

By the way, looking at the sulfur print obtained from the steel material, there are many small particles as noise even in the part without segregation, and in the part of the actual center segregation, there are particles with a slightly higher concentration than the noise particles. . Therefore, when a density histogram is created from an image including center segregation, the center segregation component is distributed with a deviation from the average image component as shown in FIG. In this case, as in Japanese Patent Laid-Open No. 63-19543, when a histogram is created and a threshold value is obtained from the average value of the density and the standard deviation, the average particle component that is noise and the central segregation can be distinguished. Concentration (Lth in Figure 6b)
In addition, since the threshold value cannot be set accurately, there is a high possibility that the center segregation is erroneously identified.

Therefore, a second object of the present invention is to accurately separate a component to be identified and a noise component in image processing to eliminate an identification error.

[Means for Solving the Problem (1)] Information on an image such as a sulfur print is binarized by using a certain density as a threshold value for the change in shading.
The image of the defective area can be roughly extracted. However, since the individual defect areas appearing as particles on the binary image are a part of the defect, if the defect is identified only by that, an error is large and accurate evaluation cannot be performed. That is, in general, only one surface is inspected, and even in the case of one defect particle (including a crack) that is actually continuous when viewed from another surface in the depth direction that does not appear on that surface, it is inspected. On the surface, it is often divided into multiple particles, or the density is partially reduced and it is often interrupted in the middle on a binary image.For example, all particles with an area of a predetermined value or less are excluded as noise particles. Then, there is a high probability that the actual segregation or crack will be overlooked.

Therefore, in the first aspect of the invention, binary image information that binary-codes the grayscale change of each part of the image is generated, and a non-defect candidate area among a plurality of defect candidate areas appearing in the binary image information is created. , The defect candidate pixels are interpolated based on the distribution state of the densities of the respective pixels and the peripheral pixels, and the plurality of defect candidate regions are connected as one region, and for each region identified by the connection, , Identify defects based on at least one of their size and shape.

[Operation (1)] For example, when one linear cracking defect is thin in the middle, the density on the image is reduced, and when it is separated and extracted into two particle regions on the binary image, the binary value When the density distribution on the image before conversion is examined, a density distribution satisfying a specific condition is found between the two extracted particle regions.
Therefore, if a pixel satisfying the condition is regarded as a pixel indicating a part of an effective defect and is interpolated between two extracted particle regions on the binary image, the two particle regions are connected. Since it is extracted and extracted as one defect area,
The error at the time of binarization, that is, the discontinuity of the defect is eliminated, and the defect can be detected with high accuracy.

Note that the first invention is applied in the internal crack detection processing in the embodiments described later.

[Means for Solving the Problem (2)] In the second invention, a plurality of defect candidates appearing in the binary image information are generated by generating binary image information representing the density change of each part of the image in a binary manner. An area is extracted, and in the peripheral pixel area with respect to the extracted defect candidate area, in a predetermined range in which the density distribution in the rotation direction is relatively large, the other defect candidate area having the shortest distance is classified into the same group. The defect is registered as a defect candidate area, and the defect is identified based on at least one of the entire area and shape of the defect candidate area group belonging to each group.

[Operation (2)] For example, in the case of V segregation, each of the impurity particles exists within a range of a predetermined angle in which the direction thereof is relatively narrow, but a large number of small impurity particles are collected without forming large particles. Often forms one segregation region. Therefore, in this case, if a large number of particles extracted on the binarized image are correlated with their orientations and distances and those having a high correlation are classified into the same group, each group has one impurity particle. It can be regarded as an area. That is, even if the area of each particle is small, if the group has a large area, the group can be regarded as a candidate for the impurity particles that form V segregation.

The second invention is applied in the V segregation detection process in the embodiments described later.

[Means for Solving the Problem (3)] In the third invention, the image information captured by the image capturing means is processed, and the defect having a relatively high frequency is included in the image.
The average density of the density distributions of the pixels forming the image is detected for the area, and the standard deviation of the density distribution of the pixels forming the image is detected for the second area in which the frequency of defects on the image is relatively low. Then, a threshold density is set based on the average density and the standard deviation, and at least the first area is binarized by dividing the change in the density of the image information by the threshold density. Generate value image information.

[Operation (3)] When detecting the center segregation, as described above, the concentration of the segregation component is higher in the portion where the center segregation is present than in the other portions, so that the variation in the concentration is large, and the segregation is obtained in that portion. Since there can be large variations in standard deviation, even if the threshold density is set based on both the standard deviation and the average density,
Only the central segregation component cannot be accurately extracted on the binary image information.

However, in the third invention, the standard deviation detected from the second region in which the frequency of inclusion of defects is relatively low is used. That is, since there is no segregation in the second region,
The concentration distribution is concentrated in a relatively narrow concentration range as shown in FIG. 6a, for example, and the distribution is irrelevant to the state of segregation. Therefore, by using the standard deviation in this case, the optimum threshold value can be determined. When the obtained image density is binarized, the segregation component and the other noise components can be accurately distinguished.

The third invention is applied in the center segregation detection process in the embodiments described later.

[Embodiment] FIG. 1 shows the appearance of a defect detection apparatus embodying the present invention.

Referring to FIG. 1, the main body of the apparatus includes an image processing apparatus 1
00, CRT display 200, keyboard 210,
A mouse 220 and a printer 230 are provided. The image reading unit 300 includes a video camera 310 for imaging.
And lamps 320 and 330 for illumination are provided. A sulfa print 400 is arranged at a position facing the video camera 310.

The sulfaprint 400 is produced by a known sulfaprinting method from a cross-section Sp of a part of a slab as shown in FIG. Specifically, in this example, the length (height) in the thickness direction of the slab having the cross-section Sp is 30 in this example.
The width of 0 mm and Sp is 150 mm.

An example of the image of the sulfa print is shown in FIG. In FIG. 3, the Su side corresponds to the upper surface of the slab, and the S side corresponds to the lower surface of the slab. The sulfur print shown in FIG. 3 is obtained by transferring the distribution state of the sulfur component on the surface of the Sp cross section (the surface indicated by hatching) to the paper surface using a chemical.

Referring to this sulfaprint, the vicinity of the axis XC (A
A few black particles appear inside the area), which is the central segregation. Also, around the center segregation (C region, D
In the region), there are clusters of elongated particles oriented in the tilt direction, which is V segregation. Further, in the area E, elongated black stripes extending in the lateral direction in FIG. 3 are seen, which are internal cracks.

The apparatus shown in FIG. 1 inputs an image of a sulfur print as shown in FIG. 3 and performs image processing to automatically identify center segregation, V segregation and internal cracks, and evaluate the quality of the slab.

FIG. 4 shows an electrical configuration of the device shown in FIG. Referring to FIG. 4, an image processing device 100 of this device includes a microprocessor (MPU) 101, a floating point arithmetic unit (FPH) 102, a main memory 103, an image processor 104, an image memory 105, an external storage device 106, CRT controller 107, keyboard IFC
(Interface) 108, mouse IFC 109, printer IFC 110, video camera IFC 111 and host IFC 112 are provided, and the CRT display 200, keyboard 210, mouse 220, printer 230, video camera and A lighting lamp is connected.

The image captured by the video camera 310 is divided into vertical and horizontal 1024 pixel regions in the video camera IFC 111 in synchronization with the image scanning timing, and the image density of each pixel is converted into 8-bit digital data. To be done. The image memory 105 is a video camera I
It has a capacity capable of simultaneously storing one or more frames of a 1024 × 1024 pixel screen with 8-bit gradation output from the FC. Normal processing is performed by the microprocessor 101.
However, the image processor 104 is configured to perform the time-consuming complicated repetitive processing peculiar to the image processing.

Although not shown, the unit of the image processor 104 includes a data conversion processor, a filtering processor, and a labeling processor.

In this example, either the 8-bit image information generated by the video camera IFC 111 or the image information obtained by binarizing the density generated by the process described below is used to transmit the CRT display 200 via the CRT controller 107.
Displayed above. The keyboard 210 is a computer (1
01), and the mouse 220 is used to specify coordinates. The result of the quality evaluation process is output to the printer 230.

FIG. 5a shows an outline of the flow of defect detection processing. 5a
The outline of the processing will be described with reference to the drawings. First, step A
In 1, a known method is used to create a sulfaprint 400 of a slab to be evaluated. In step A2, a reference image for calibrating variations in the sensitivity of the image reading system (video camera and illumination) is set at the image reading position of the image reading unit 300 instead of the sulfa print 400. The position of the video camera 310 is adjusted so that the resolution of the image, that is, the size of each pixel on the current image has a predetermined value.

In step A4, the created sulfa print is set in the image reading unit 300 as shown in FIG. The image of the sulfa print 400 is now displayed on the CRT display 200.

In the subroutine of step A6, processing for detecting the center segregation is performed, and in the subroutine of step A7, V
A process for detecting segregation is performed, and a process for detecting internal cracks is performed at the sub routine of step A8.
In step A9, the information of the defects detected in steps 6, A7 and A8, that is, the result of the quality evaluation is displayed in the printer 230
Output to.

Hereinafter, the specific operation of the apparatus will be described for each of the center segregation process of step A6, the V segregation process of step A7, and the internal cracking process of step A8 shown in FIG. 5a.

First, the center segregation process will be described. The detailed contents of this process are shown in FIG. 5b. The processing of each step will be described with reference to FIG. 5b.

In step B1, the area of the image to be processed is set. The area of the image processed by the center segregation processing is shown in Fig. 3A.
A region (hereinafter referred to as A region) and a region indicated by B (hereinafter referred to as B region). Specifically,
Area A is a square area of ± 10 mm in the x direction and 150 mm in the y direction (entire sulfaprint) from the position of the center axis XC, and area B is 35 mm in the opposite direction from the center axis XC to the x direction.
± 10mm in the X direction and 150 in the y direction centered on the position
It is a rectangular area of mm (entire sulfaprint).

The reason for defining each area as described above is as follows. A
In the x direction (10 mm) of the area, center segregation occurs in the range of ± 10 mm, which is the center of any steel grade, and the range for quality evaluation is within that range in operation. Has been set. In the y direction (150 mm) of the area A, this is done so that the sulfur print used for quality evaluation has a constant size regardless of the steel type. The area B has the same condition as the area to be evaluated, and thus the same area size is set.

The area A in step B2 and the area B in step B3
The image information of bit gradation (density 0 to 255) is read into a predetermined position on the image memory 105 for image processing.

In step B6, the standard deviation σB of the density of the B area is obtained based on the image density data of the B area.

In addition, when the histogram of the B area is represented in a graph,
It becomes like Figure a.

In step B7, the image information of the area A is processed to obtain the average density μA of the area. It should be noted that the histogram of the image information of the area A is shown in a graph, for example, as shown in FIG. 6b.

Referring now to FIG. 3, on the sulfaprint,
In addition to the components of segregation and cracking of relatively large particles, very small particles (black dots), which are not particularly problematic in terms of quality, are present as noise. The center segregation is highly likely to occur in the area A, but the segregation or cracking is extremely unlikely in the area B due to the nature of continuous casting. That is, the particles appearing in the B area are noise components, and the particles appearing in the A area include both noise components and segregation components. Looking at the density distribution in the A area, as shown in FIG. 6b, the center segregation component has a slightly higher density than the other noise components, and if divided by the threshold density indicated by Lth, It can be seen that the noise component and the center segregation component can be separated.

Such a threshold density Lth is equal to the standard deviation of the area A and A
It seems that it can be set accurately based on the average concentration of the region, but in reality, the amount of sulfur component differs depending on the type of slab, the concentration difference of segregation noise changes depending on the detection target, and depending on the quality. The amount of center segregation also changes. Therefore, if the threshold concentration for extracting the segregation is set only by the concentration information of the area A where the segregation is to be recognized and set, the threshold value itself is affected by the amount of the segregation, and the extraction of the central segregation is performed. There is a risk of making an error.
For example, when the standard deviation of the area A is used, when the segregation is large, the segregation is small, and when the segregation is small, the segregation is large.

Therefore, in this embodiment, as shown in the next step B8, the threshold density Lth is set to the standard deviation σB of the B area including only the noise component and the average density μA of the A area having the center segregation.
And set accordingly. In addition, n is a predetermined constant. Therefore, in this embodiment, even if a large change occurs in the component of the center segregation, the threshold concentration Lth
Is always set at the boundary between the noise component and the center segregation component.

In step B9, the threshold density Lth set in step B8 is used to binary-identify the change in density of the image information of 8-bit gradation in the area A to generate binary image information. .

That is, information indicating whether each pixel is black (high density) or white (low density) is generated for each pixel in the x direction and the y direction. When this information is displayed two-dimensionally in association with the arrangement of pixels on the image, it becomes as shown in FIG. 7b, for example. In FIG. 7b, the hatched pixels are black pixels, and all other pixels are white pixels.

In step B10, the binary image information obtained in the process of step B9 is processed to extract the impurity particles in the area A. That is, a set of black pixels adjacent to each other, for example, the seventh
Each of PT1 and PT2 in the diagram b is extracted as one particle.

In step B11, the area of each of the particles (PT1, PT2, ...) Extracted in step B10 is calculated. In practice, the number of black pixels forming each particle is calculated, and the numerical value is set as the area S.

In step B12, the particle size (radius) of each of the particles extracted in step B10 is obtained. Actually, the shapes of particles are various, so the particle size is calculated by using the square root of particle size = (area S / π) as an approximate expression using the area of each particle calculated in step B11.

In step B13, the average value of the particle sizes of the particles excluding the fine particles that can be regarded as noise is calculated.

In step B14, among the extracted particles, the particles having the largest diameter up to the fifth are registered as important segregated particle components.

In step B15, the total area of each particle obtained in step B11 is calculated.

In step B16, the total area (ΣS) of the segregation components obtained in step B15 and the area (15
The area ratio of the center segregation is calculated from (0 × 20 mm).

Area ratio = ((ΣS × Km) / (150 × 20)) × 100 (%) Note that Km is a constant for converting the number of pixels into an actual area.

In step A9 shown in FIG. 5a, the result of the above-described center segregation processing is that the average value of the particle size, the area and the particle size of the registered five particles, and the area ratio are output by the printer, and the area If the ratio or the average particle size is equal to or larger than a predetermined threshold value, it is determined that there is a problem in the quality of the slab, and information indicating that the slab has a defective center segregation is output to the printer.

Next, the V segregation process will be described. The detailed contents of this process are shown in FIGS. 5c and 5d. The processing of each step will be described with reference to the drawings.

In step C1, the area of the image to be processed is set. The area of the image processed by the V segregation processing is shown in C in FIG.
The area in the range indicated by (hereinafter referred to as C area) and the area in the range indicated by D (hereinafter referred to as D area). Specifically,
The width of the C area in the x direction above the A area (x direction) is 40 m.
The length in the m and y directions is 150 mm (total length of the sulfa print), and the D area is below the area (opposite x direction).
Is a region having a width in the x direction of 40 mm and a length in the y direction of 150 mm. V segregation occurs within the C and D regions. In this V segregation processing, the same processing is performed for each of the C area and the D area.

8 in the C area in step C2 and the D area in step C3
The image information of bit gradation is read into a predetermined position on the image memory 105 for image processing.

In step C6, the standard deviation σ of each of the C area and the D area
Is detected, and the average density μ of each region is detected in step C7.

At step C8, the threshold density Lth is set. In this example, the threshold value of the C area is the standard deviation C of the C area.
It is set according to the average density of the area, and the threshold value of the D area is
It is set according to the standard deviation of the D area and the average density of the D area. n is a predetermined constant.

In step C9, using the threshold density Lth set in step C8, the change in density of the image information of 8-bit gradation in each of the C area and the D area is binary-identified, and the binary value is changed. Generate image information. That is, information indicating whether each pixel is black (high density) or white (low density) is generated for each pixel in the x direction and the y direction. When this information is two-dimensionally displayed in association with the arrangement of pixels on the image, it becomes as shown in FIG. 8a, for example. In FIG. 8b, the hatched pixels are black pixels, and all other pixels are white pixels.

In step C10, the binary image information obtained in the process of step C9 is processed to extract the impurity particles in the C region and the D region. That is, a set of black pixels adjacent to each other,
For example, PT31, PT32, PT3 in FIG. 8a
3, PT4, ... are each extracted as one particle.

In step C11, the area of each of the particles extracted in step C10 is obtained. Actually, the number of black pixels forming each particle is calculated, and the numerical value is set as the area S.

In step C12, the barycentric position of each of the particles extracted in step C10 is obtained. Assuming that the pixel position of the center of gravity in the x direction is the pixel position of the center of gravity in the y direction, the position of the center of gravity of each particle is obtained by the following equation (1).

= (Σxi) / n, = (Σyi) / n (1) where x
i: x-coordinates of pixels in particles (i = 1 to n) yi: y-coordinates of pixels in particles (i = 1 to n) n: number of black pixels forming particles In step C13, detection in step C10 is performed. One particle is extracted from the large number of particles thus obtained. The processing from this step to step D13 in FIG. 5d is repeatedly executed, and is executed for each of all the particles.

In step C14, the area of one particle extracted in step C13 is examined. For example, the particles PT shown in FIG. 8a
When the area is relatively large as in No. 4, the particle alone is regarded as a candidate for particles constituting V segregation, and the process proceeds to the next step C15, otherwise the particle is small. , Step D1 in Fig. 5d. n2
Is a predetermined constant.

In step C15, the covariance γ of the n pixels (xi, yi) forming the extracted particles is obtained. The covariance γ is a parameter that can be used to identify whether or not the particles composed of those pixels are elongated, and the closer the elongated shape is, the closer it is to 1 and the more rounded it is, the closer it is to 0. In this case, γ is calculated by the following equation (2).

In step C16, the covariance γ obtained in step C15
Identify the size of. When γ is larger than the predetermined value n3, the particles are relatively elongated and there is a high possibility that they are particles that form V segregation. Therefore, the process proceeds to the next step C17 to register the particles as effective particles, and otherwise. To proceed to step D1 in FIG. 5d. In step C18, determining the moment m _11, m ₂₀ and m ₀₂ of the registration particles as active particles in step C17. Each moment is calculated by the following equation (3).

m ₁₁ = Σ (xi −) (yi−) m ₂₀ = Σ (xi−) ² m ₀₂ = Σ (yi−) ² (3) At step C19, the moment obtained at step C18 is used. The extending direction of the registered elongated particles is obtained as an angle θ with respect to the x axis. This angle is calculated by the following equation (4).

θ = (1/2) ・ tan ^-1 [m ₁₁ / (m ₂₀ -m ₀₂ )]…
(4) For particles that have been considered to have a relatively small area in step C14 or particles that have a relatively round shape in step C16, refer to step D in FIG. 5d.
The process starting from 1 is applied.

In step D1, the size of the area of the particles to be identified is identified. The reference value n4 in this case is a predetermined constant and is a value smaller than n2 in step C14. If the area of the particles is smaller than n4, that is, if the particles are very small particles that cannot be regarded as candidates for V-segregated particles, step D
If the area of the particle is n4 or more, the process proceeds to step D2.

In step D2, particles having an area of n4 or more are regarded as particle elements capable of forming V segregated particles by a plurality of particles, and are registered.

In step D3, the particle or particle element in the vicinity of the particle element registered in step D2 is searched for. That is, V
In the case of segregation, for example, as shown in FIG. 8b, a large number of particles are unevenly distributed in a specific direction. Therefore, the correlation between particles is examined, and those having a strong correlation are connected.

Specifically, first, within the range of the radius R (corresponding to 7 mm in this example) with the center of gravity (or center) of the particle of interest as the center, the sum of the densities of all pixels in each angle θi direction (on the binary image The number of black pixels may be used), and a density histogram in the angular direction as shown in FIG. 8c is created. In the case of V segregation, since the distribution of segregated particles depends on a specific direction, a peak appears at an angle on the histogram.
That is, in the direction of the peak angle θo with respect to the particle of interest,
There will be particles. Therefore, in step D3, the particle is searched for within the range of the angle θo ± Δθ with respect to the particle of interest.

In step D5, among the found particles, the particle closest to the target particle is regarded as a particle having a strong correlation with the target particle, and the particle is registered in the same particle group as the target particle.

For example, when particles P0, P1, P2, and P3 exist as shown in FIG. 8d, the center of gravity Cp of the first particle of interest P0 is
The particle having the closest (center of gravity) within the range of θo ± Δθ with respect to o is P1, so P1 is registered in the same group as the particle P0.

In step D6, the particle of interest is moved to the particle element newly registered as an element of the group in the immediately preceding step D5, and the process returns to step D3. Therefore, when there are many particle elements that can be registered in the same group, the processes of steps D3, D4, D5 and D6 are repeatedly executed many times. In the case of shifting from step D6 to D3, if the angle in the concentration distribution direction of the new target particle is greatly different from the angle θo in the concentration distribution direction detected in the previous target particle, the particle elements of the same group are Is considered to be nonexistent.

For example, when particles P0, P1, P2, and P3 are present as shown in FIG. 8d, P1 is first registered in the group by the above process, and then P2 is correlated with P1.
Are registered in the group, and then the particle P3 is registered in the group in correlation with P2. That is, in this case, the 8th
As shown in the figure, four particles P0, P1, P2, and P3 are registered in one group.

When the detection of all particles that can be registered as elements of the same group as the particle element registered in step D2 is completed, the process proceeds from step D4 to step D7.

In step D7, the total number of particles classified into the same group as the particle element registered in step D2 is checked. If there are three or more, then proceed to step D8,
Otherwise, it proceeds to step D13.

In step D8, the total sum of the areas of three or more particle elements divided into the same group is obtained, and the size of the sum is checked.
n5 is a predetermined constant.

If the total area of the particle group is relatively large, the process proceeds to step D9, and the entire group is regarded as an effective particle group that constitutes V segregation and is registered.

In step D10, the center of gravity of all the particle groups belonging to one group registered in step 9 is obtained. This process is the same as step C12 described above. In step D11,
The moment of the entire particle group belonging to one group registered in step D9 is obtained. This calculation process is the same as in step C18 described above.

In step D12, the tilting direction (angle) of all the particle groups belonging to one group registered in step D9 is obtained. This calculation process is the same as in step C19 described above.

The above processing is executed for each of all the particles extracted in step C10. When the processing is completed for all particles, the process proceeds from step D13 to step D14.

In step 14, the angles of each of the V segregation components (particles and particle groups) obtained in the above process (step C
19 or the value obtained in D12) is taken as the angle of V segregation. Since the directions of the segregated particles in the regions C and D are opposite to each other, the angles of the particles in the regions C and D are averaged separately.

In step D15, the total area of the V segregation constituent elements is calculated, and the area ratio, that is, the ratio of the area of the V segregation particle portion to the area of the C region and the D region is calculated.

In step A9 of FIG. 5a, the result (angle and area ratio) of the V segregation processing in step A7 is output to the printer, and if the area ratio is not less than a predetermined value, V
Information indicating that there is a poor segregation is output to the printer.

Next, the internal cracking process will be described. The detailed contents of this processing are shown in FIGS. 5e, 5f and 5g. The processing of each step will be described with reference to the drawings. In step E1,
Set the area of the image to be processed. The areas of the image to be processed by the internal cracking processing are the area in the range indicated by E in FIG. Specifically, the E region is a region having a width of 65 mm in the x direction and a length of 150 mm in the y direction on the upper side (x direction) of the A region, and F region
The region has a width of 65 in the x direction on the lower side of the A region (the direction opposite to x).
This is a region having a length in the mm and y directions of 150 mm. It is in these regions (especially region E) that internal cracks are likely to occur. In this internal cracking process, the same process is performed for each of the C area and the D area.

In step E2, E area, and in step E3, F area 8
The image information of bit gradation is read into a predetermined position on the image memory 105 for image processing.

In step E6, the standard deviation σ of each of the E region and the F region
Is detected, and in step E7, the average density μ of each region is detected.

At step E8, the threshold density Lth is set. In this example, the threshold value of the E region is set according to the standard deviation of the E region and the average density of the E region, and the threshold value of the F region is the standard deviation of the F region and the average density of the F region. Set accordingly. n is a predetermined constant.

In step E9, the threshold density Lth set in step E8 is used to binary-identify a change in the density of 8-bit gradation image information in each of the E region and the F region, and a binary image is obtained. Generate information.

That is, information indicating whether each pixel is black (high density) or white (low density) is generated for each pixel in the x direction and the y direction. When this information is displayed two-dimensionally in association with the arrangement of pixels on the image, it becomes as shown in FIG. 9a, for example. In FIG. 9a, the hatched pixels are black pixels, and the other pixels are all white pixels.

In step E10, E obtained in the process of step E9
A process based on the binary image information of the area is performed to extract a crack area candidate in the E area. This process is complicated and will be described in detail later. In step E11, a process based on the binary image information of the F region obtained in the process of step E9 is performed to extract a crack region candidate in the F region. This process is similar to the process of step E10.
In step E12, the length X in the x direction and the length Y in the y direction are obtained for each of the crack regions in the E region and the F region extracted in step E10 or E11. In particular,
Let X be the length from the minimum pixel to the maximum x coordinate of the pixels forming the cracked region, and let Y be the length from the minimum pixel to the maximum y coordinate of the pixels constituting the cracked region. (See FIG. 9e) At step E13, the crack gradient, that is, Y / X is calculated.

In step E14, the number of pixels in the cracked area is calculated and the area thereof is obtained.

In step E15, the area ratio of the crack region is calculated.

At step E16, it is determined whether the internal cracking occurs.

Specifically, in this example, the length of the cracked region is 3 mm or more, and the angle of inclination of the cracked region is ± with respect to the x axis.
When all the conditions of being within 30 degrees and having a relatively elongated shape are satisfied, it is determined as an internal crack. Regarding the angle of cracking, such conditions are set because cracking occurs only in that direction due to the nature of casting. Regarding the shape, when the larger one of the lengths X and Y of the cracked region is L and the area of the cracked region is S,
If the condition of S / L ² <0.2 is satisfied, it is considered to be elongated.

Next, the E area connection processing in step E10 will be described. This content is shown in FIGS. 5f and 5g. First, a description will be given with reference to FIG. 5f.

In step F1, pixels are sequentially scanned for the binary image information generated in step E9, and black pixels (those having high density) are searched for. Example 9 of binary image information
Shown in Figure a. The scanning sequence is such that one line is scanned in the x direction, and when the scanning is completed, the coordinate in the y direction is moved to the next pixel position and the scanning in the x direction is performed again.

The processing after step F2 is executed for all the black pixels in the E area and the F area.

When a black pixel is found in step F1, the process proceeds to step F3 via step F2, and 1 is set in the flag Fdir indicating the direction.

In step F4, the state of the flag Fdir is identified. If the flag Fdir is 1, then proceed to step F5,
Otherwise, it proceeds to step G1 in FIG. 5g.

Here, the relative pixel positions are defined as shown in FIG. 9b. The nine pixels P _{11 to} P ₃₃ shown in FIG. 9b are used.
The central pixel P ₂₂ is set as the target pixel.

In steps F5, F6, and F7, the density of each pixel is read by referring to the 8-bit gradation image information of the black pixel (pixel of interest) searched in step F1 and the pixels in the vicinity of the black pixel (the pixel of interest), and the identification thereof is performed. To do. That is, in step F5, the density of the pixel of interest P _{22 and} the density of the pixel P _{32 on} the right are compared. In step F6, the difference between the density of the pixel P _{32 and} the density of the target pixel P ₂₂ is determined. n6 is a constant, σ
Is the standard deviation of the concentration. In step F7, the pixel P ₃₂
Check the magnitude of the concentration of. The reference value in this case is μ-n7
-Σ (μ is the average value of the concentration, n7 is a constant).

If any one of the conditions of steps F5, F6 and F7 is satisfied, the process proceeds to step F8, and if not, the process proceeds to F18.

In step F8, a density histogram in the rotation direction around the pixel of interest is created. Specifically, within the distance of 9 pixels from the pixel of interest,
As shown in the figure, the sum of the densities is obtained for each of the angles of 10 degrees, and a histogram is created. As a result, for example, a histogram as shown in FIG. 9d is obtained. In the case of internal cracking of the slab, the cracking direction is in the x direction as shown in FIG. 3, so in the histogram including information on internal cracking, as shown in FIG. 9d, the angular region R1 close to the x axis is shown. And R2 (region of ± 30 degrees with respect to the x-axis) has a distribution larger than the densities of other angle regions R3 and R4.

In step F9, the maximum sum of the densities of the nine pixels is extracted from the angular region R1, and the density R1m thereof is compared with the threshold value THa. This threshold THa is obtained from the following equation (5).

THa = (μ + n8 · σ) × 9 (5) That is, since the comparison is made by the total sum of the densities of nine pixels, the normal threshold value is multiplied by nine and used.

Here, μ is the average concentration of the E or F region, σ is the standard deviation of the concentration of the E or F region, and n8 is a constant.

In step F10, the angular region R2 having the maximum total density is extracted and its density R2m is compared with the threshold value THa.

In step F11, the maximum density R in the angular region R1
1m is compared with the threshold THb. The threshold THb is
It is obtained from the following equation (6).

THb = [(R3 + R4) / 2] + n9 · σ (6) where R3: average density of the angular area R3 R4: average density of the angular area R4 n9: constant σ: standard of the density of E area E or F area In the deviation step F12, the maximum density R in the angle region R2
2m is compared with the threshold THb.

When the conditions of steps F9 and F11 are satisfied or when the conditions of steps F10 and F12 are satisfied, the process proceeds to step F13, and otherwise, the process proceeds to step 17.

In step F13, the pixel having the highest density among the three pixels P ₃₁ , P ₃₂ , and P ₃₃ on the right side of the pixel of interest P ₂₂ is selected as Pm, and in step F14, the pixel of interest is moved to the pixel of Pm. .

In step F15, the content of the counter CNp is incremented. The content of CNp is cleared to 0 when step F3 is executed.

In step F16, the pixel of Pm is stored, and step F
Go back to 5. The processing of F5 to F16 is repeatedly executed while moving the target pixel until the conditions are not satisfied. Then, when these conditions are not satisfied, the process proceeds to step F17, and when the counter CNp is three times or more, the pixel of interest Pm is set as the connected pixel in step F18. As a result, even if a pixel whose density is lower than the threshold when binarizing the density and is considered to be a white pixel is likely to be part of the crack, it is set as a black pixel. Handle it. As a result, for example, in the case of one crack region which has a low concentration in the central portion and is divided into two crack regions and binarized, black pixels,
That is, since the connected pixels are interpolated, the two crack areas are connected, and the original shape of the entire crack area can be restored.

The flag Fdir indicates the moving direction of the pixel of interest. When it is 1, it moves in the x direction, and when it is 0, it moves in the opposite direction of x (minus direction).

After the execution of step F18, the flag Fdir is 0, so after step F4, step G in FIG.
Go to 1.

In steps G1, G2, and G3, the density of each pixel is read by referring to the 8-bit gradation image information of the black pixel (pixel of interest) searched in step F1 and the pixels in the vicinity thereof, and the identification thereof is performed. To do. That is, in step G1, the density of the target pixel P ₂₂ is compared with the density of the pixel P _{12 on} the left. In step G2, the difference between the density of the pixel P _{12 and} the density of the target pixel P ₂₂ is determined. In step G3, the density of the pixel P ₁₂ is checked. The reference value in this case is μ-n7 · σ.

If any one of the conditions of steps G1, G2, and G3 is satisfied, the process proceeds to step G4, and if not, the process returns to step F1 to search for the next black pixel.

In step G4, the same processing as step F8 described above is performed, and the flow advances to step G5.

The respective means G5, G6, G7 and G8 respectively perform the same processing as the steps F9, F10, F11 and F12 described above.

If the conditions of steps G5 and G7 are satisfied or the conditions of steps G6 and G8 are satisfied, the process proceeds to step G9, and if not, the process proceeds to step F1.

In step G9, the three pixels P on the left side of the target pixel P ₂₂ are
_{Of 11} , ₁₁ , P ₁₂ , and P ₁₃ , the one with the highest concentration is P
The pixel of interest is moved to the pixel of Pm in step G10.

At step G11, the content of the counter CNp is incremented. Note that the content of CNp is cleared to 0 when moving from step F4 to G1.

In step G12, the pixel of Pm is stored, and step G
Return to 1. The processing from G1 to G12 is repeatedly executed while moving the target pixel until the condition is not satisfied. Then, when the condition is not satisfied, the process proceeds to step G13, and when the counter CNp is three times or more, the pixel of interest Pm is set as the connected pixel in step G14. As a result, even if a pixel whose density is lower than the threshold when binarizing the density and is considered to be a white pixel is likely to be part of the crack, it is set as a black pixel. Handle it. As a result, for example, in the case of one crack region which has a low concentration in the central portion and is divided into two crack regions and binarized, black pixels,
That is, since the connected pixels are interpolated, the two crack areas are connected, and the original shape of the entire crack area can be restored.

The processes of F1 to G14 are repeatedly executed until the above processes are completed for all the black pixels obtained by binarizing the density.

An example of the result of the above processing is shown in FIG. 9e. In FIG. 9e, the hatched pixels are pixels that are regarded as black pixels (high-density pixels above a threshold value) by binarizing the density, and the other pixels are pixels that are regarded as white pixels. is there. In addition, each pixel marked with a circle represents a connected pixel interpolated by the above-described connecting process. In the example of FIG. 9e, two black pixel regions identified as a result of binarization are connected to one by the connecting pixel generated by the connecting process. In this case, the length of the crack area is identified as X in the x direction and Y in the y direction.

The contents of the F region connecting process shown in step E11 of FIG. 5e are the same as the above E region connecting process, and only the region to be processed, the standard deviation to be used, and the average density are different.

In the above embodiment, the image of the sulfur print is read using a video camera that electrically scans both in the vertical direction and in the horizontal direction. However, for example, a line sensor that performs one-dimensional solid-state scanning and a mechanical sensor are used. The image may be read using an image scanner in combination with another scanning drive mechanism.

Further, in the embodiment, the defect information is obtained from the image of the sulfa print, but the image may be captured by other means as long as the image includes similar defect information.

〔effect〕

As described above, according to the present invention, by the image processing device,
Since information of quality evaluation regarding defects is automatically generated, ambiguity, individual difference, variation, etc., which occur when humans perform sensory evaluation, are eliminated, and a highly accurate quality evaluation result can be obtained. Moreover, in the present invention, since there is a special device for extracting the pixels corresponding to the defect, the ability to discriminate the defect from the non-defective part is very high as compared with the conventional simple density binarization processing. .

[Brief description of drawings]

FIG. 1 is a perspective view showing the external appearance of one type of defect detection apparatus embodying the present invention. FIG. 2 is a perspective view showing a cast piece to be inspected. FIG. 3 is a plan view showing an example of a sulfa print obtained from a cast piece. FIG. 4 is a block diagram showing a configuration of an electric circuit of the apparatus shown in FIG. FIG. 5a is a flow chart showing the flow of a quality evaluation process performed using the apparatus shown in FIG. FIGS. 5b, 5c, 5d, 5e, 5f, and 5g show the operation of the image processing apparatus 100, and FIG.
The figure shows the center segregation process of step A6 of FIG. 5a, and FIGS. 5c and 5d show the V segregation process of step A7 of FIG. 5a.
5e, 5f and 5g are flow charts showing the contents of the internal cracking process of step A8 of FIG. 5a, respectively. FIGS. 6a and 6b are graphs showing the density distributions in the images of the B area and the A area on the sulfur print, respectively. FIG. 7a is a plan view showing an area A on the sulfa print,
FIG. 7b is a plan view showing a part of the information obtained by reading the image of FIG. 7a and binarizing the density. FIGS. 8a and 8b are plan views showing a part of the information obtained by reading the image of the area C on the sulfa print and binarizing the density, and FIG. 8c shows the rotation direction around the particle PO of FIG. 8b. FIG. 8d is a plan view showing the positional relationship between the search region and the center of gravity of each particle, and FIG. 8e is a plan view showing the positional relationship of information on a series of particles registered in the same group. FIG. 9a is a plan view showing a part of the information obtained by reading the image of the E region on the sulfa print and binarizing the density.
FIG. 9b is a plan view showing the positional relationship of each pixel in the vicinity of the target pixel, FIG. 9c is a plan view showing the division of the target pixel and the area in the rotation direction, and FIG. 9d is a density distribution in the rotation direction around the target pixel. The graph shown, FIG. 9e, is a plan view showing the information after the information of FIG. 9a has been concatenated. 100: image processing apparatus 101: microprocessor 102: floating point arithmetic unit 103: main memory 104: image processor 105: image memory 200: CRT display 210: keyboard 220: mouse 230: printer 300: image reading unit 310: video camera 320 , 330: Lamp 400: Sulfur print

 ─────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-63-19543 (JP, A) JP-A-53-88650 (JP, A) JP-A-52-120231 (JP, A)

Claims (3)

[Claims] 1. An image pickup means for picking up optical image information showing a change in structure of a cross section of a steel material to be inspected; a binary image which binary-processes image information picked up by the image pickup means and binary-codes a change in shading of each part of the image. Information is generated, and in the non-defect area between the plurality of defect candidate areas appearing in the binary image information, the pixel of the defect candidate is based on the distribution state of the density of each pixel and the surrounding pixels. Image processing means for interpolating the plurality of defect candidate areas as one area and connecting each of the plurality of defect candidate areas as one area, and identifying a defect based on at least one of the size and shape of each area identified by the connection. And an output unit for outputting information on the defect identified by the image processing unit; 2. An image pickup means for picking up optical image information showing a change in structure of a cross section of a steel material to be inspected; a binary image which binary-processes the image information picked up by the image pickup means and binary-codes the light and shade change of each part of the image. Information is generated, a plurality of defect candidate areas appearing in the binary image information are extracted, and in the peripheral pixel area with respect to the extracted defect candidate areas, the density distribution in the rotation direction is within a predetermined range in a relatively large direction, Of the other defect candidate areas, the one having the shortest distance is registered as a defect candidate area of the same group, and the defect is identified based on at least one of the entire area and shape of the defect candidate area groups belonging to each group. An image processing unit; and an output unit that outputs information on the defect identified by the image processing unit; 3. An image pickup means for picking up optical image information showing a structural change of a cross section of a steel material to be inspected; image information picked up by said image pickup means is processed, and a defect having a relatively high frequency is included in the image. The average density of the density distributions of the pixels forming the image is detected for the area, and the standard deviation of the density distribution of the pixels forming the image is detected for the second area in which the frequency of defects on the image is relatively low. Then, a threshold density is set based on the average density and the standard deviation, and at least the first area is binarized by dividing the change in the density of the image information by the threshold density. Image processing means for generating value image information and identifying a defect based on at least one of the size and shape of each of the defect candidate areas appearing in the binary image information; and the image processing means Missing identified And outputs the information output means; comprises, quality evaluation unit of the slab cross-section.